Coseismic Landslide Area Prediction Using Generalised Additive Model: A Case Study of the 2013 Minxian Earthquake

ABSTRACT This study aims to establish a regional model for predicting seismic landslide areas. Using the 2013 Minxian earthquake‐induced landslide database as the research foundation, mathematical statistics and GIS techniques were applied to predict landslide areas through the Generalised Additive Model (GAM). The study area was divided into slope units using r.slopeunits, with these units serving as the basis for landslide area prediction. The influencing factors such as elevation, slope angle, profile curvature, distance to seismogenic fault (Dis2fault), distance to epicentre (Dis2epicenter), peak ground acceleration (PGA), distance to rivers (Dis2rivers) and lithology were selected for analysis. The predicted landslide areas for different slope units were calculated using the GAM and then compared with actual landslide distribution. The results show that slope angle and Dis2fault have a more significant impact on the spatial distribution of landslide areas compared with other influencing factors. Slope angle shows a positive correlation with landslide occurrence; the landslide area increases with the rise of slope angle. For the Dis2fault, the actual distribution of landslides shows that most landslides primarily occur on both sides of the seismogenic fault, indicating a significant effect of the fault on landslide distribution. Otherwise, our modelling result indicates that the predicted landslide areas align well with the actual distribution. However, a notable tailing effect was observed in regions with either very small or large landslide areas. Specifically, in slope units with less developed landslide areas, the model tended to overestimate the size, whereas in areas with more extensive landslides, the model tended to underestimate the actual area.